C#: local variable initializers and multiline call arguments can be misclassified as phantom function definitions (`content`, `FromSeconds`)

[Widthdom]

Summary

While reviewing main..HEAD with the locally built binary (dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll), I found a separate C# symbol-extraction false positive:

• a local variable declaration like const string content = "hello"; is indexed as a phantom function content
• a multiline assertion/call argument like elapsed < TimeSpan.FromSeconds(10), is indexed as a phantom function FromSeconds with bogus return_type = "elapsed <"

This pollutes symbols / outline and can make definition point at non-definitions.

Minimal repro

CDIDX='dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll'

TMPDIR=$(mktemp -d /tmp/cdidx-csharp-phantom-fn.XXXXXX)
cat > "$TMPDIR/Repro.cs" <<'EOF'
using System;

namespace Demo;

public class Repro
{
    public void M(TimeSpan elapsed)
    {
        const string content = "hello";
        Assert.True(
            elapsed < TimeSpan.FromSeconds(10),
            $"x {elapsed.TotalSeconds:F2}");
    }
}

public static class Assert
{
    public static void True(bool condition, string message) { }
}
EOF

"$CDIDX" index "$TMPDIR" --json >/dev/null
DB="$TMPDIR/.cdidx/codeindex.db"
"$CDIDX" symbols --db "$DB" --json --lang csharp

Observed extra rows:

{"kind":"function","name":"content","line":8,"signature":"const string content = \"hello\";","return_type":"string"}
{"kind":"function","name":"FromSeconds","line":10,"signature":"elapsed < TimeSpan.FromSeconds(10),","return_type":"elapsed <"}

Expected:

• only the real symbols (namespace Demo, class Repro, function M, class Assert, function True)
• no function content
• no function FromSeconds

Why it matters

This is not just cosmetic noise:

• outline of C# test files becomes misleading
• definition content / definition FromSeconds can point at statements or expressions instead of real definitions
• AI consumers using symbol lists for orientation get false structure inside method bodies

I first noticed it on this repo's own tests/CodeIndex.Tests/SymbolExtractorTests.cs, where outline surfaced both a phantom content symbol from a fixture local and a phantom FromSeconds symbol from an Assert.True(...) argument.

Suspected direction

The C# method/function regex path is still accepting some statement-shaped lines once extraction is inside a method body.

Likely hardening points:

• reject local variable declarations (const string content = ..., var x = ..., typed locals)
• reject expression fragments / argument lines that only contain a call site (elapsed < TimeSpan.FromSeconds(10),)
• add explicit regression tests for both shapes in tests/CodeIndex.Tests/SymbolExtractorTests.cs

Scope

• src/CodeIndex/Indexer/SymbolExtractor.cs
• tests/CodeIndex.Tests/SymbolExtractorTests.cs

Environment

• Reproduced on macOS arm64
• Binary: local build from current HEAD via dotnet ./src/CodeIndex/bin/Debug/net8.0/cdidx.dll
• Discovered during adversarial review of fix-issue361

[Widthdom]

Started working on this in branch fix-issue496. I’m reproducing the false-positive C# symbol extraction cases with the locally built binary and will update this issue if the scope changes.

[Widthdom]

Continuing work for this in branch fix-issue496-782, together with related issue #782.

[Widthdom]

Continuing work for this in branch fix-issue496-782-785, together with related issues #782 and #785.